Artificial intelligence (AI) is everywhere – or so you would believe if you listened to the hype.
But in truth, we’ve barely scratched the surface in exploring what smart software can deliver. What is often labelled AI today is actually machine learning, a process whereby software starts out with a defined set of rules, computes through reams of big data and begins to learn or create new rules.
Although this is an impressive process, it’s not really artificial intelligence. In fact, to call it AI is a bit like comparing a dinghy to the Titanic.
Computers excel over human beings in being able to instantly process an extraordinary amount of data. That’s why even something as simple as automation can look like intelligence. Automated software has the ability to go through vast databases looking for whatever set of variables it’s been instructed to before any human has even got out of bed.
The arrival of big data, and its ongoing exponential growth, has been critical in driving advances in automated software and machine learning. It’s not just the amount of data that has to be processed, but also the types of data and where and how it’s inputted.
We’re moving from a world of static data to a dynamic one. Examples of data that tends to stay fairly static are your name, address, date of birth and telephone number. They will change, but over time. You can get a human being to very accurately code and match these types of data attributes.
When you move to a dynamic world, you still use static data but add things like your geolocation, which can change by the minute. Dynamic data can also include your transaction history, websites you’ve visited, social posts you’ve made, and the device ID you have.
Using a human being to go through and write the code for all that data isn’t possible. You have to start with an initial rule set written by a human that enables a machine to take all the data inputs and start writing its own code and rules on an automated basis, slowly over time.
For example, machine learning is proving to be a valuable resource in identifying malware. While hundreds of thousands of new malware files pop up every day, the code behind these files (often available to bad actors for sale on the dark web) changes very little between iterations. A machine learning algorithm can be given the base code for a wide range of malware files and then ‘learn’ the deviations to keep identifying new threats as they occur.
While data is dynamic, changing in real time, sadly so too are criminal elements attempting to breach databases, defraud businesses or attack cyber systems. Machine learning is a big step towards tackling this. Delivering software that can change itself in real time, spotting new patterns and responding instantly, instead of sending a report to a human who then might spend days, or weeks, coding a response.
For machine learning to work, the machine needs a huge amount of data to cycle through in the first place in order to “learn”. This inevitably means that initially the machine might get things wrong. Once you get to the level of data input that a human can’t check, you have to rely on the machine to get it right, and if it’s not, you may not be able to tell.
For example, regulators are highly focused on preventing things like money laundering. At the moment, we can be very clear about why we say yes to an individual. We can say, that individual matched x, y, z database and we checked x, y, z identity documents, therefore, we did everything we could to verify that individual and approved the loan. In a subsequent compliance investigation, that would be a clean bill of health.
But in the new world of machine learning, an audit could reveal that a check actually did fail, but the machine overrode the decision because it had built a rule itself to do so in certain conditions, and approved the load. That’s going to make for some interesting conversations.
Because we need ever-more intelligent software to process our growing databases, legislation and regulation will follow and spur better and better decision-making systems. But even the most learned machines can’t tell us yet if we will ever see true artificial intelligence.
Author: Gareth Stephens is Head of Product at GBG.
Sources:  http://www.kaspersky.com/about/news/virus/2014/Kaspersky-Lab-is-Detecting-325000-New-Malicious-Files-Every-Day